US20130173232A1

US20130173232A1 - Method for determining the course of the road for a motor vehicle

Info

Publication number: US20130173232A1
Application number: US13/641,748
Authority: US
Inventors: Urban Meis; Christoph Wiedemann; Wladimir klein; Christoph Brüggemann
Original assignee: Conti Temic Microelectronic GmbH
Current assignee: Conti Temic Microelectronic GmbH
Priority date: 2010-04-20
Filing date: 2011-03-31
Publication date: 2013-07-04
Also published as: US10776634B2; DE102010020984A1; DE112011100146A5; WO2011131165A1; JP2013530435A; EP2561419B1; EP2561419A1

Abstract

A method for determining the course of the road for a moving motor vehicle having a surroundings sensor system. Sensor data generated by the surroundings sensor system are evaluated to detect lane-relevant features. A lane model having at least one lane model parameter that determines the course of the lane is generated for the road, structures of at least one distance range that are parallel to the road are detected in the sensor data, the tangential direction of at least the one structure that is parallel to the road is determined, and the value of the tangential direction of the structure that is parallel to the road is adopted as the value of the direction of the tangent line at the point of contact with the structure that is parallel to the road to determine at least the one lane model parameter by predictive estimation in the lane model.

Description

CROSS REFERENCE TO RELATED APPLICATIONS

This application is the U.S. National Phase Application of PCT/DE2011/000359, filed Mar. 31, 2011, which claims priority to European Patent Application No. 10401058.2, filed Apr. 20, 2010 and German Patent Application No. 10 2010 020 984.8, filed May 19, 2010, the contents of such applications being incorporated by reference herein.

FIELD OF THE INVENTION

The invention relates to a method for determining the course of the road for a moving motor vehicle with at least one surroundings sensor system according to a method for determining the course of the road for a moving motor vehicle with at least one surroundings sensor system, wherein the sensor data generated by the surroundings sensor system that is directed toward the road are evaluated in order to detect lane-relevant features. Furthermore, the invention relates to a motor vehicle with a device for carrying out the inventive method.

BACKGROUND OF THE INVENTION

Lane detection is an important component of driver assistance systems that are used, e.g., for longitudinal control or for lane keeping.
Known lane detection systems are based on assigning the detected structures to the side of the lane and the distance thereof from the middle of the lane. The difficulty of said assignment increases with an increasing distance of the measuring system from the structures. Furthermore, the description of the detected lane usually refers to the position of the vehicle and does not provide any explicit information about the far range. The usual lane model uses the curvature and the curvature change of a clothoid model to describe the further course of the road, wherein the estimated values are an averaging relative to the ego-position of the vehicle. Therefore, the known methods are not capable of determining the exact point of transition between two clothoid segments since roads are created by stringing clothoid segments together. For example, the transition from a bend to a straight line cannot be determined precisely, which results in an unnatural control behavior of a vehicle lateral control system of a driver assistance system at these points of transition.
According to U.S. Pat. No. 6,718,259 B1, which is incorporated by reference, this problem is solved by supplying the surroundings data generated by a sensor system to a filter bank made up of several Kalman filters, wherein the lane model is based on a clothoid model in which the road region in front of the vehicle is subdivided into a near range (up to a distance d1) and a far range (distance from d1 to d2) having different clothoid parameters, wherein a continuous transition between these two ranges is assumed. Thus, a point of transition between two clothoid segments of the road is not estimated, but the transition point is assumed to be at distance d1. Each of the individual Kalman filters of the filter bank is adapted to a lane model, which lane models differ from each other with respect to the position of the transition point relative to distance d1. Each Kalman filter of this filter bank provides an estimate of the lane model parameters of the respective model, wherein each of these estimated values is subsequently weighted with a weighting value that corresponds to the probability of the occurrence of the respective model. The weighted output data are merged.

SUMMARY OF THE INVENTION

The disadvantage of this known method according to U.S. Pat. No. 6,718,259 B1 consists in the high complexity of the filter bank made up of Kalman filters and results in long runtimes. The transition point is not detected but filtered out so that the model error remains present.
An aspect of the invention provides a method of the type mentioned at the beginning that is improved as against the known method and that can detect the transition points of the course of the curvature of the road and thus correctly model the course of the lane right into the far range.
According to an aspect of the invention, a method for determining the course of the road for a moving motor vehicle with at least one surroundings sensor system in which the sensor data generated by the surroundings sensor system that is directed toward the road are evaluated in order to detect lane-relevant features is characterized in that

a lane model having at least one lane model parameter that determines the course of the lane is generated for the road,
structures of at least one distance range that are parallel to the road are detected in the sensor data,
the tangential direction of at least the one structure that is parallel to the road is determined, and
the value of the tangential direction of the structure that is parallel to the road is adopted as the value of the direction of the tangent line at the point of contact with the structure that is parallel to the road in order to determine at least the one lane model parameter by means of a predictive estimation method in the lane model.

According to this inventive method, all structures that are parallel to the road, such as lane markings, curbstones, structures on the roadside, crash barriers and the like, are used to estimate the course of the lane without having to additionally introduce filtering quantities for the estimation method. Furthermore, this inventive method is not dependent on the lateral distance of the structures that are parallel to the road nor is it necessary to estimate a deviation from the middle of the lane for the motor vehicle so that the number of the degrees of freedom is small which results in an estimation method that is more robust, more efficient with respect to runtime and less error-prone. Thus, the course of the lane can be modeled over the entire considered distance range by means of a lane model with extremely precise estimated values, wherein all lane models that result in continuously differentiable functions are suitable. Such lane models include, aside from a clothoid road model, the circular-arc model, the double-clothoid model and the spline model, for example.
Furthermore, the inventive method is also not dependent on the type of the surroundings sensor system used. Thus, an image-based sensor (e.g., a camera) as well as a radar sensor, a lidar sensor or a digital map with GPS within a navigation system or a combination thereof may be used.
Finally, the inventive method can also be used to predict lane-related system boundaries that can be advantageously used in driver assistance systems. For example, a narrowing bend in an approach road can be detected in advance so that a driver assistance system can prepare the driver of the vehicle for a crossing of a system boundary or indicate a bend radius located 60 m ahead of the vehicle or a minimal bend radius.
According to an advantageous further development of the invention, structures that are parallel to the road are determined from the sensor data by means of an edge detection method by determining and evaluating gray-scale gradients. To this end, a Hough transform is preferably used, by means of which shapes, e.g., straight lines or circles made up of edge points, can be detected. In order to preclude false structures, a criterion of parallelism is advantageously introduced, by means of which criterion it is possible to detect structures that are parallel to the road and that are located on both sides of the road and in front of the motor vehicle and at the same distance from the motor vehicle, and by means of which criterion it is also possible to preclude non-parallel structures.
Furthermore, according to an advantageous realization of the invention, the predictive estimation of the lane model parameter is performed by means of a Kalman filter model that is approximated, e.g., by means of a third-order polynomial known in the art. Of course, estimation methods that are comparable with the Kalman filter model may be used as well.
Finally, according to a further development of the invention, the advantageous properties of the inventive method are shown by the fact that the lane width of the road, the deviation of the motor vehicle from the middle of the lane and the yaw angle of the motor vehicle are determined from the sensor data that cover the near range of the road and the further course of the lane is determined by means of the at least one generated lane model parameter.
This shows that the inventive method results in a reduction of the phase space because the inventive method is not dependent on the lane width or on the deviation of the vehicle from the middle of the lane, whereby the method is made more robust and more efficient with respect to runtime. This partitioning of the phase space into a near range and a far range prevents changes of the width of the lane, e.g., lane widening or exit ramps, from negatively influencing the lane model parameters. The lane width and the deviation of the vehicle from the middle of the lane can be restricted to the relevant near range and do not have to be estimated and averaged for the entire measuring range.
The inventive method can be advantageously used in a vehicle having a driver assistance system, e.g., within a longitudinal control system, a lane keeping system or a lane departure warning system.

BRIEF DESCRIPTION OF THE DRAWINGS

In the following, the invention will be explained in greater detail with reference to the single attached FIG. 1 that shows structures that are parallel to the road and structures that are not parallel to the road in order to explain the inventive method, which structures are shown in a road coordinate system of a road.

DETAILED DESCRIPTION

A surroundings sensor system that may be a camera, a radar or lidar sensor as well as a digital map with GPS (e.g., within a navigation system) provides sensor data for an evaluation unit by means of which the inventive method is carried out.
Said sensor data are used to carry out a feature extraction by the detection of structures that are parallel to the road, such as crash barriers, verges, lane markings, curbstones or other demarcating structures that are parallel to the road.
For example, an edge detection method is used to this end during video processing, by means of which method the contours of such structures that are parallel to the road can be detected, wherein an edge is regarded as a change of the gray-scale values, i.e., the gray-scale gradients are determined and evaluated by means of a Hough transform.
To this end, the sensor data are subdivided into search areas having surroundings-related horizontal and vertical boundaries, wherein the lane detected in the near range is used for the vertical boundaries.
A clothoid model having clothoid segments is used as a lane model. Each clothoid segment is described by an initial curvature c₀and a curvature change c₁and approximated in the x-y coordinate system of the road by means of a third-order polynomial:
$l (x) = y = \frac{c_{0}}{2} x^{2} + \frac{c_{1}}{6} x^{3}$
From the sensor data, an edge image is extracted that corresponds to the structure that extends tangentially to the middle of the lane. Furthermore, said edge image has a slope m as a tangent line at a point x in front of the vehicle. Said slope m is determined by means of a Hough transform.
In the lane model, the slope m of said tangent line is adopted as the value of the direction of the tangent line at the point of contact with the structure that is parallel to the road and that represents the tangent line.
Such a situation is shown in FIG. 1 that is shown in the coordinate system of the road. The road is shown as a two-lane road with a lane 1 that is demarcated by a verge 2 and a center line 3. The middle of the lane 4 of the lane 1 extends through the origin of coordinates and tangentially to the x-axis. The verge 2 represents a clothoid segment. The end of said clothoid segment is denoted by reference numeral 7 that may also represent a point of transition to the next road segment.
Said FIG. 1 shows, in position x₁, an individual center line 3 as a structure that is parallel to the road, which structure was extracted from image data of a road scene that was recorded by means of a camera. A tangent line T1 with a slope m₁that extends parallel to the middle of the lane (indicated by an arrow) is determined in the centroid of said individual center line 3.
Furthermore, FIG. 1 shows, in position x₂, a further structure 5 that is parallel to the road, which structure 5 consists of several objects that are arranged in a straight line and detected as stationary radar targets by means of, e.g., a radar sensor. In the centroid of said structure, a tangent line T2 with a slope m₂extends parallel to the middle of the lane 4 (indicated by an arrow).
Finally, FIG. 1 shows a further straight-line verge structure 6. However, said structure 6 does not extend parallel to the road.
The sensor-data-based measurement of the slopes m_iand of the associated positions x₁is used to estimate the lane model parameters c₀and c₁of the clothoid model by means of a Kalman filter that has the following form in the x_i-y_icoordinate system of the vehicle in the above-mentioned approximation:
$l (x_{i}) = y_{i} = y_{0 L / R} + θ x_{i} + \frac{c_{0}}{2} x_{i}^{2} + \frac{c_{1}}{6} x_{i}^{3},$
wherein y_0LIRis the deviation (offset) of the vehicle from the middle of the lane to the left or to the right and θ is the yaw angle of the vehicle.
The associated measurement equation in the Kalman filter model is obtained by differentiating the above equation l(x_i) and equating with the slope m_i:
$m_{i} = θ + c_{0} x_{i} + \frac{c_{1}}{2} x_{i}^{2} .$
By means of this measurement equation, the estimated lane model parameters c₀and c₁are corrected in order to determine the further course of the lane (represented by segments K1 and K2 in FIG. 1). Thus, the course of the lane can be modeled over the entire considered distance range, in particular over the far range.
Aside from the single-clothoid model described herein, a double-clothoid model or a node-based spline model may be used as a lane model.
It is also possible to determine the curvature c₀in the considered distance segment by means of the slope m prior to estimating the lane model parameters c₀and c₁. To this end, the course of the lane is approximated in each distance segment by the parabola
$l (x) = \frac{c_{0}}{2} x^{2},$
wherein c₀is the curvature in the x-position. The slope m of detected structures that are parallel to the road (which slope m is determined from the sensor data) corresponds to the slope of the parabola in the same x-position and is given by the first derivative, i.e., c₀x=m , so that the curvature c₀is:
$c_{0} = \frac{m}{x} .$
In order to preclude false structures (i.e., structures that are not parallel to the road), a two-dimensional Hough method may be used to search the detected distance range for pairs of straight lines that extend parallel to each other on opposite sides of the road. By means of such as criterion of parallelism, straight lines like the verge structures 6 shown in FIG. 1 can be precluded.
In the inventive method described herein, the detection of structures that are parallel to the road does not require any knowledge of the lateral distance of said structures, whereby it is not necessary to estimate the lateral offset of the vehicle, either.
The parameters y_0LIR(distance or offset of the vehicle from the middle of the lane) to be determined in the near range and the lane width can be determined by means of individual measurements, which results in an advantageous partitioning of the phase space with the offset and the lane width from the near range and the lane model parameters c₀and c₁in the far range. The yaw angle θ is also determined in the near-range partition. Such a partitioning of the phase space prevents changes of the width of the lane, e.g., lane widening or exit ramps, from negatively influencing the lane model parameters, i.e., it is not dependent thereon and thus causes the degrees of freedom to be reduced, whereby the filter system (the Kalman filter in this exemplary embodiment) is made more robust, i.e., less error-prone, and more efficient with respect to runtime and results in more precise estimated values.
Since the course of the lane detected in the near range is extrapolated into the far range by means of the inventive method, it is not necessary to average and estimate the offset and the lane width in the far range.
In order to preclude false straight-line verge structures (shown in FIG. 1) whose straight lines G do not extend parallel to the road in the centroid, a criterion of parallelism is used, as described above, when evaluating the sensor data. According to said criterion, only structures that are parallel to the road and parallel to each other and that extend in the same position left of and right of the lane are determined, for example.
Each individual measurement (tangent line) can be converted into a piece of curvature information in the x-position. If one looks at the individual tangent measurements sequentially in the x-direction, the measurements exhibit a deviation from the filtered lane model from clothoid transition point 7 up (FIG. 1), whereby said transition point 7 can be determined. Thus, a lateral vehicle control that corresponds to the natural driving behavior can be performed within a driver assistance system.
Within a driver assistance system, it is also possible to predict the reaching of lane-related system boundaries. For example, a narrowing bend in an approach road can be detected in advance so that a driver assistance system can prepare the driver of the vehicle for a crossing of a system boundary or indicate a bend radius located 60 m ahead of the vehicle or a minimal bend radius.
The inventive method demonstrates that the additional use of structures that are parallel to the road (such as markings, curbstones, structures on the roadside, and crash barriers) for the predictive determination of bend radii is possible without having to extend the filters.

Claims

1.-8. (canceled)

9. A method for determining the course of the road for a moving motor vehicle with at least one surroundings sensor system, wherein sensor data generated by the surroundings sensor system that is directed toward the road are evaluated in order to detect lane-relevant features,

wherein

a lane model having at least one lane model parameter that determines the course of the lane is generated for the road,

structures of at least one distance range that are parallel to the road are detected in the sensor data,

a tangential direction of at least the one structure that is parallel to the road is determined, and

a value of the tangential direction of the structure that is parallel to the road is adopted as the value of the direction of the tangent line at the point of contact with the structure that is parallel to the road in order to determine at least the one lane model parameter by a predictive estimation method in the lane model.

10. The method according to claim 9, wherein continuously differentiable segments are used as the lane model.

11. The method according to claim 10, wherein continuously differentiable segments are used as the lane model within a clothoid road model, a circular-arc model or a spline model.

12. The method according to claim 11, wherein structures that are parallel to the road are determined by an edge detection method by determining and evaluating gray-scale gradients.

13. The method according to claim 12, wherein the gray-scale gradients are evaluated by a Hough transform.

14. The method according to claim 12, wherein by a criterion of parallelism, structures that are parallel to the road and that are located in front of the motor vehicle and at the same distance from the motor vehicle are detected and non-parallel structures are precluded.

15. The method according to claim 11, wherein by a criterion of parallelism, structures that are parallel to the road and that are located in front of the motor vehicle and at the same distance from the motor vehicle are detected and non-parallel structures are precluded.

16. The method according to claim 9, wherein the predictive estimation of the lane model parameter is performed by a Kalman filter model or a comparable estimation method.

17. The method according to claim 9, wherein a lane width of the road, a deviation of the motor vehicle from the middle of the lane and a yaw angle of the motor vehicle are determined from the sensor data that cover the near range of the road and the further course of the lane is determined by the at least one generated lane model parameter.

18. The method according to claim 9, wherein structures that are parallel to the road are determined by an edge detection method by determining and evaluating gray-scale gradients.

19. A vehicle with a driver assistance system, for carrying out a method for determining the course of the road for a moving motor vehicle with at least one surroundings sensor system, wherein sensor data generated by the surroundings sensor system that is directed toward the road are evaluated in order to detect lane-relevant features, wherein